Abstract: The task of generating human-scene interaction involves multiple fields such as computer vision, computer graphics, and robot control. This aims to utilize deep learning algorithms to model and learn the interaction relationships between humans and scenes from a large amount of interaction motion data, generating various interaction motions of humans with indoor scenes or objects within them, including obstacle avoidance navigation, human-chair interaction, and object grasping, etc. Compared to traditional physics-based simulation methods, data-driven interaction motion generation break away from the dependence on physical simulation engines, possessing higher computational efficiency and stronger generalization capabilities. These methods have broad application prospects in fields such as game design, film production, and human-computer interaction. However, current research on human-scene interaction motion generation technology has not yet formed a systematic summary. This work systematically organizes and elucidates the core advancements in current human-scene interaction motion generation technologies. Firstly, it explains the data representation methods for 3D humans and scenes; based on this, it systematically summarizes different types of interaction tasks and their technical challenges, detailing the core features of relevant benchmark datasets and the evaluation metric systems; finally, it summarizes the limitations of current technical routes and analyzes the breakthrough directions and potential development paths for future research.